How Can I Solve NaN Output Problem

Hı, ı have a code P(i,j) is float and it has value 0-1 range , normally if we calculate log function must be equal positive value with initial minus. But this return to NaN what is the wrong

function entropy=ent(glcm)
    P=glcm/sum(glcm(:));
    entropy=0;
    for i=1:8
        for j=1:8
            entropy=entropy+ P(i,j)*(-log(P(i,j));
        end
    end
end

[per isakson]

I fail to reproduce the error you report (on R2018)

>> ent(rand(8))
ans =
       3.9484
>> ent(rand(8))
ans =
       3.9371
>> ent(rand(8))
ans =
       4.0212 

What input to the function do you use?

[Furkan b]

I want to calculate texture features like entropy using GLCM matrix this function part of my all code but it does not work. it return NaN.

[Roger J]

What is the input to the function?

If glcm matrix is all zeros, then the sum is zero, and that creates NaN in P, and finally the returned entropy is NaN.

Also, don't name the variable "entropy", there is a common function named entropy(..).

Answer 1

[Steven Lord]

As Roger Jestes stated, if any element of P is exactly 0, you compute 0*(-log(0)) which is 0*inf which results in a NaN. This is the second of the indeterminant forms listed in this section of the Wikipedia page for NaN.

>> 0*(-log(0))

ans =

   NaN

NaN plus anything is still NaN so once you add NaN to your entropy value once that's what it will remain.

[Furkan b]

Thank you so much